As far as I can tell, our chat does not currently allow Chatgpt-style “personalization”, where the model remembers important facts about you from your chat history and uses them as context in future conversations. I believe the term for this in Librechat is “automatic memory extraction” and shouldn’t be too hard to implement. See here for more info: https://www.librechat.ai/docs/configuration/librechat_yaml/object_structure/memory
Why does it matter?
I have a friend whose interest in joining the co-op is conditional on this feature, so that’s reason enough for me. I’ve never used it myself but it does seem to be a popular feature on ChatGPT from my anecdotal experience.
Who is it for?
Members who use the chat.
How might it work?
From the librechat docs (see link above):
Automatic memory extraction, where a separate memory agent reviews every exchange and saves facts without being asked, is a separate opt-in: set agent.enabled: true and give it a model.
Anything else?
I can’t think of any reasons why someone would object to allowing this feature but I would love to hear of any objections or concerns if there are any.
## What's the idea?
As far as I can tell, our chat does not currently allow Chatgpt-style “personalization”, where the model remembers important facts about you from your chat history and uses them as context in future conversations. I believe the term for this in Librechat is “automatic memory extraction” and shouldn’t be too hard to implement. See here for more info: https://www.librechat.ai/docs/configuration/librechat_yaml/object_structure/memory
## Why does it matter?
I have a friend whose interest in joining the co-op is conditional on this feature, so that’s reason enough for me. I’ve never used it myself but it does seem to be a popular feature on ChatGPT from my anecdotal experience.
## Who is it for?
Members who use the chat.
## How might it work?
From the librechat docs (see link above):
> Automatic memory extraction, where a separate memory agent reviews every exchange and saves facts without being asked, is a separate opt-in: set agent.enabled: true and give it a model.
## Anything else?
I can’t think of any reasons why someone would object to allowing this feature but I would love to hear of any objections or concerns if there are any.
Thanks for raising this, and for digging up the right LibreChat feature — you found the correct mechanism.
What's possible on our chat: the installed LibreChat (v0.8.8) supports a memory block in its config, members already get a per-user on/off switch for memories, and there's a view/delete path so people can see what's stored. MongoDB is already running behind our chat, so the storage requirement is met. It's configuration and copy, not a rebuild.
Three options:
Manual memory only (my recommendation to start). Members can ask the chat to remember something; no automatic extraction. Most of the benefit, none of the profiling.
Full ChatGPT-style automatic extraction, opt-in. A separate memory agent reviews each exchange and saves facts. Needs a model choice and a price tag, because it adds a call per exchange — roughly doubling token use for anyone who enables it.
Wait until members ask for it.
Two things I'd want us to be explicit about before switching on option 2, since they're the kind of detail that bites later:
Cost: extraction uses tokens. If it bills the member's key, their allowance drains faster and they should be told; if the co-op absorbs it, that's a new line item.
Who sees the exchange: the memory agent reads prompts and replies. If we want to keep the architectural-privacy story intact for extraction, that agent should be a Tinfoil TEE model rather than a policy-based one. Memories themselves sit in our database in plaintext, like chat history already does.
So: yes, doable; I'd ship manual memory first, default off, with a visible toggle, and add automatic extraction as an opt-in once people ask for it. Happy to write the copy and wire whichever option the group prefers.
Thanks for raising this, and for digging up the right LibreChat feature — you found the correct mechanism.
**What's possible on our chat:** the installed LibreChat (v0.8.8) supports a `memory` block in its config, members already get a per-user on/off switch for memories, and there's a view/delete path so people can see what's stored. MongoDB is already running behind our chat, so the storage requirement is met. It's configuration and copy, not a rebuild.
**Three options:**
1. **Manual memory only (my recommendation to start).** Members can ask the chat to remember something; no automatic extraction. Most of the benefit, none of the profiling.
2. **Full ChatGPT-style automatic extraction, opt-in.** A separate memory agent reviews each exchange and saves facts. Needs a model choice and a price tag, because it adds a call per exchange — roughly doubling token use for anyone who enables it.
3. **Wait** until members ask for it.
**Two things I'd want us to be explicit about before switching on option 2,** since they're the kind of detail that bites later:
- **Cost:** extraction uses tokens. If it bills the member's key, their allowance drains faster and they should be told; if the co-op absorbs it, that's a new line item.
- **Who sees the exchange:** the memory agent reads prompts and replies. If we want to keep the architectural-privacy story intact for extraction, that agent should be a Tinfoil TEE model rather than a policy-based one. Memories themselves sit in our database in plaintext, like chat history already does.
So: yes, doable; I'd ship **manual memory first, default off, with a visible toggle**, and add automatic extraction as an opt-in once people ask for it. Happy to write the copy and wire whichever option the group prefers.
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
What's the idea?
As far as I can tell, our chat does not currently allow Chatgpt-style “personalization”, where the model remembers important facts about you from your chat history and uses them as context in future conversations. I believe the term for this in Librechat is “automatic memory extraction” and shouldn’t be too hard to implement. See here for more info: https://www.librechat.ai/docs/configuration/librechat_yaml/object_structure/memory
Why does it matter?
I have a friend whose interest in joining the co-op is conditional on this feature, so that’s reason enough for me. I’ve never used it myself but it does seem to be a popular feature on ChatGPT from my anecdotal experience.
Who is it for?
Members who use the chat.
How might it work?
From the librechat docs (see link above):
Anything else?
I can’t think of any reasons why someone would object to allowing this feature but I would love to hear of any objections or concerns if there are any.
Thanks for raising this, and for digging up the right LibreChat feature — you found the correct mechanism.
What's possible on our chat: the installed LibreChat (v0.8.8) supports a
memoryblock in its config, members already get a per-user on/off switch for memories, and there's a view/delete path so people can see what's stored. MongoDB is already running behind our chat, so the storage requirement is met. It's configuration and copy, not a rebuild.Three options:
Two things I'd want us to be explicit about before switching on option 2, since they're the kind of detail that bites later:
So: yes, doable; I'd ship manual memory first, default off, with a visible toggle, and add automatic extraction as an opt-in once people ask for it. Happy to write the copy and wire whichever option the group prefers.